Overview
The connection between OPD and RL; Apply Empirical Bayes in OPD.
Mathematical Formulation
Consider the expected reward
\[ J(\pi)=\mathbb{E}_{\tau\sim\pi}[R(\tau)]. \]
We want to find
\[ \pi^*=\arg\max_{\pi}J(\pi). \]
Open Questions
- What assumptions are necessary?
- Can we derive a theoretical guarantee?